A technique for preserving lighting and detail consistency: generate a continuous video sequence rather than a series of separate still images, because consecutive frames of one video generation inherently match each other while independently generated stills do not. Addresses the within-shot half of Continuity Is the Creator's Responsibility (Not the Model's).
Once a consistent video sequence is generated, individual frames can be pulled out and used as standalone still assets: pause the video at the desired moment and screenshot it, rather than generating that still separately (and risking a mismatch with its neighbors).
This is the general form of the technique used for Blocking Lock via Position Reference's position log — extraction works for any still needed from an already-consistent sequence, not just positional references.
When several look-alike shots are needed from one moment, generate one continuous video covering that range and extract frames from it for use as still assets, instead of generating each still independently.
A manual technique for carrying continuity between shots when no direct handoff exists: screenshot the final frame of the previous generated shot and feed that screenshot in as the start frame for the next shot's generation. This keeps lighting, framing, and character state continuous across a cut even when the tool doesn't natively chain shots.
A manual implementation of anchor-frame continuity: after a shot finishes generating, screenshot its final frame directly from the output player and feed that screenshot in as the start-frame/reference image for the next shot's generation. This chains shots frame-to-frame without needing the tool to natively support "continue from last frame" — it works around that gap using whatever frame the model happened to end on.
Из тем: Unsorted, Holding Continuity Across Independently Generated Shots